2002 Fiscal Year Final Research Report Summary
The trial production of generalized solver for optimization problems on HTC environments
| Project/Area Number |
13680511
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
社会システム工学
|
|---|
| Research Institution | Tokyo University of Agriculture and Technology |
|---|
Principal Investigator |
SHINANO Yuji Tokyo Univ. of A&T, Computer Science, Assistant Prof., 工学部, 講師 (00297623)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
IKEDA Satoshi Tokyo Univ. of A&T, Computer Science, Assistant Prof., 工学部, 助手 (70282796)
FUJIE Tetsuya Kobe Univ. of Commerce, Management Science, Assistant Prof., 商経学部, 助手 (40305678)
|
|---|
| Project Period (FY) |
2001 – 2002
|
| Keywords | Combinatorial optimization / Branch-and-bound algorithms / HTC / PC cluster |
|---|
| Research Abstract |
First of all, based on "The research purpose and research plan" indicated in the grant application, I show the targets at the time of the application. The targets within the research period for two years are as follows.
(1) The trial production of a system which implemented general parallel branch-and-bound algorithms on the HTC environments where 1000 or more PCs and workstations are connected with Internet or LAN (In its design, it is afford to operate general-purpose meta-heuristic solver in parallel).
(2) Having a clear grasp of the problems when building a generalized solver on HTC environments
(3) Having a grasp of limitations for using the generalized solver in the HTC environments
We have been developed PUBB2 (Parallelization Utility for Branch-and-Bound algorithms version 2) in this research period. About (1) and (2), it was attained through development of PUBB2 except for the point that execution by 1000 or more sets of the HTC environment was not actually performed. In PUBB2, im
… More
plementing a few functions which are forced to override in its framework, enable to build a new layer of a new execution environment. Moreover, an algorithm of load balancing for parallel and distribution execution can be implemented 〓〓exibly by developing new specialized classes which are derived from ProblemManager class in PUBB2. Although the code for a layer of 〓RID environment was underdeveloped, using the software tool we built, a new framework itself can be developed flexibly and easily.
About (3) of the above target, we tried to parallelize ILOG CPLEX MIP Optimizer which is commercial based solver for a mixed integer problem using PUBB2 framework. The ILOG CPLEX MIP Optimizer is regarded as one of the fastest and the most robust MIP Optimizer in the world. It have been reflected not only the theoretically new research results, but also it is the sophisticated program with refined memory management etc. Therefore, even the parallel system for GRID etc. does not usually have the same performance as a CPLEX run on a single computer. The parallelization utility should parallelize the latest products. Then, as one of the evaluation of our prototype system, we verified whether if our prototype of our general framework can be parallelize the MIP Optimizer efficiently or not.
The main product of this research is having developed the prototype system. Therefore, Reference Manual automatically generated from developed program source code is included in the final report. Since the research period was spent on development, although there was no refereed paper published, the results are summarized in two papers that are included in the final report. One of the papers has been accepted. Less
|
Research Products
(6 results)
